Did the Clean Air Act Cause the Remarkable Decline in ...ceepr.mit.edu › files › papers ›...
Transcript of Did the Clean Air Act Cause the Remarkable Decline in ...ceepr.mit.edu › files › papers ›...
Center For Energy and Environmental Policy Research
Did the Clean Air Act Cause the Remarkable Decline in Sulfur Dioxide Concentrations?*
Michael Greenstone
Reprint Series Number 196*Reprinted from Journal of Environmental Economics and Management, Vol. 47, No. 3, pp. 585-611, 2004, with kind permission from Elsevier. All rights reserved.
The MIT Center for Energy and Environmental Policy Research (CEEPR) is a joint center of the Department of Economics, the MIT Energy Initiative, and the Alfred P. Sloan School of Management. The CEEPR encourages and supports policy research on topics of interest to the public and private sectors in the U.S. and internationally.
The views experessed herein are those of the authors and do not necessarily refl ect those of the Massachusetts Institute of Technology.
Journal of Environmental Economics and Management 47 (2004) 585–611
Did the Clean Air Act cause the remarkable decline in sulfurdioxide concentrations?
Michael Greenstonea,b,c,�
aMIT Department of Economics, 50 Memorial Drive E52-391B, Cambridge, MA 02142-1347, USAbNBER, USA
cAmerican Bar Association, USA
Received 11 July 2002; revised 14 August 2003; accepted in revised form 1 December 2003
Abstract
Over the last three decades, ambient concentrations of sulfur dioxide (SO2) air pollution have declined byapproximately 80%. This paper tests whether the 1970 Clean Air Act and its subsequent amendmentscaused this decline. The centerpiece of this legislation is the annual assignment of all counties to SO2
nonattainment or attainment categories. Polluters face stricter regulations in nonattainment counties.There are two primary findings. First, regulators pay little attention to the statutory selection rule in theirassignment of the SO2 nonattainment designations. Second, SO2 nonattainment status is associated withmodest reductions in SO2 air pollution, but a null hypothesis of zero effect generally cannot be rejected.This finding holds whether the estimated effect is obtained with linear adjustment or propensity scorematching. Overall, the evidence suggests that the nonattainment designation played a minor role in thedramatic reduction of SO2 concentrations over the last 30 years.r 2003 Elsevier Inc. All rights reserved.
JEL classification: Q25; Q28; H22; H11
Keywords: Clean Air Act; Air pollution; Sulfur Dioxide; Benefits of environmental regulation
1. Introduction
Prior to 1970, states, counties, and municipalities were principally responsible for environ-mental regulations. In one of the most significant interventions into the marketplace of the
ARTICLE IN PRESS
�Corresponding author. MIT Department of Economics, 50 Memorial Drive E52-391B, Cambridge, MA 02142-
1347, USA. Fax: +1-617-253-1330.
E-mail address: [email protected].
0095-0696/$ - see front matter r 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.jeem.2003.12.001
post-World War II era, the federal government made a dramatic foray into the regulation of airpollution with the passage of the Clean Air Act Amendments (CAAAs) and the establishment ofthe Environmental Protection Agency (EPA) in 1970. Congress further strengthened theprovisions of the CAAAs in 1977 and 1990.The CAAAs are controversial, because reliable evidence on their costs and benefits is not
readily available. For instance, there is not even a consensus on whether the CAAAs areresponsible for the dramatic improvements in air quality that have occurred in the last 30 years[6,7,11,16,24]. The fundamental problem is that there is not a valid counterfactual for what wouldhave happened to air pollution concentrations in the absence of these regulations.This paper exploits the structure of the CAAAs to determine whether this legislation is
responsible for the remarkable 80% decline in national sulfur dioxide (SO2) concentrations from0.0301 ppm in 1970 to 0.0048 ppm in 1992.1 The centerpiece of this legislation is the establishmentof the National Ambient Air Quality Standards (NAAQS)—a minimum level of air quality thatall counties are required to meet—for SO2 and six other pollutants.2 As a part of this legislation,each US county annually receives separate nonattainment or attainment regulation designationsfor each of the pollutants. The nonattainment designation is reserved for counties whose aircontains concentrations of the relevant pollutant that exceed the federal standard. Emitters of thecontrolled pollutant in nonattainment counties are subject to greater regulatory oversight thanemitters in attainment counties.This paper empirically tests whether SO2 nonattainment status caused a reduction in SO2
concentrations relative to attainment status.3 This is an important test, because the primary aimof the CAAAs was to bring all counties into compliance with the federal standards. It is importantto note from the outset though that the results are not directly informative about the CAAAs’effect on nationwide air quality, because this legislation also introduced regulations in attainmentcounties.The analysis is conducted with comprehensive county-level data files on SO2 concentrations,
nonattainment designations, and economic activity for the 1969–1997 period. Through a Freedomof Information Act request, I obtained summary information on the universe of state and nationalSO2 monitors from the EPA. Although the SO2 attainment/nonattainment status of each countyis central to federal environmental law, this is the first study to collect them (from the Code of
Federal Regulations) and use this information during this key period. The pollution and regulationdata are merged to county-level economic data from the Bureau of Economic Analysis.An immediate, yet surprising, finding is that the vast majority of nonattainment counties had
pollution concentrations below, in many cases substantially so, the federal SO2 standard. Despiteextensive communications with the EPA, I am unable to conclusively determine why they received
ARTICLE IN PRESS
1These figures are not based on a fixed set of monitors but nevertheless are illustrative of the long-run decline in SO2
concentrations.2The other pollutants are carbon monoxide, lead, nitrogen dioxide, ozone, small particulate matter, and total
suspended particulates.3A number of recent studies have employed similar econometric strategies: Becker and Henderson [3], Greenstone
[12], and Henderson [16] examine the relationship between nonattainment status and manufacturing activity;
Henderson [16] documents the relationship between ozone nonattainment status and ozone concentrations; Chay and
Greenstone [7] explore the impact of TSPs nonattainment status on TSPs concentrations and housing prices; and Chay
and Greenstone [6] assess the effects of TSPs nonattainment status on TSPs concentrations and infant mortality rates.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611586
this designation. Thus, this paper attempts to evaluate the impact of SO2 nonattainment statuswhen the selection rule is unknown. The evaluation problem is made even more difficult by thefinding that in the ‘‘pre-period’’ nonattainment counties have higher SO2 concentrations andlarger downward trends in ambient SO2 than the attainment counties.To control for the likely confounding due to these observable differences, linear regression
and propensity score matching techniques [20] are employed to estimate the effect of SO2
nonattainment status in three different 6-year periods, 1975–80, 1981–86, and 1987–92. Theanalysis reveals that matching is especially useful when the observable differences between thenonattainment and attainment counties are greatest. The results indicate that SO2 nonattainmentstatus is associated with modest reductions in SO2 air pollution, but a null hypothesis of zeroeffect generally cannot be rejected. Overall, the paper finds that the nonattainment designationplayed a minor role in the dramatic reduction of SO2 concentrations over the last three decades.The paper proceeds as follows. Section 2 describes the structure of the CAAAs and provides
some background information on sulfur dioxide pollution. Section 3 discusses the data sourcesand presents some summary statistics. Section 4 describes the research design and some of thechallenges to its validity. Section 5 presents the econometric strategy and Section 6 discusses theresults. The paper ends with brief interpretation and conclusion sections.
2. Background on the evolution of air pollution law and sulfur dioxide pollution
The ideal analysis of the relationship between air pollution and environmental regulationswould involve a controlled experiment in which the regulations are randomly assigned. In thiscase, regulatory status is independent of all other determinants of air quality. Consequently, thesubsequent differences in air quality can be causally related to the regulations. In the absence ofsuch an experiment, an appealing alternative is to find a case where the intensity of regulationdiffers across similar counties. The structure of the CAAAs may provide such an opportunity.This section describes the CAAAs and how their form may provide the opportunity to crediblyidentify the relationship between clean air regulations and sulfur dioxide concentrations. It alsoprovides a brief background on reviews sulfur dioxide pollution and its consequences.
2.1. The CAAAs and nonattainment status
The centerpiece of the CAAAs is the establishment of separate NAAQS for SO2 and other‘‘criteria’’ air pollutants, which all counties are required to meet.4 For SO2 pollution, a county isin violation of the standards if: (1) its annual mean concentration exceeds 0.03 parts per million(ppm), or (2) the second highest 24-h concentration exceeds 0.14 ppm. The stated goal of theCAAAs is to bring all counties into compliance with these standards by reducing local airpollution concentrations. To achieve this goal, stricter regulations are imposed on emitters innonattainment counties than in attainment ones.The EPA annually assigns separate nonattainment–attainment designations for each of the
criteria pollutants. The nonattainment designation is supposed to be reserved for counties whose
ARTICLE IN PRESS
4See Lave and Omenn [17] and Liroff [18] for more detailed histories of the CAAAs.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 587
actual or modeled air pollution concentration exceeds the federal ceiling. Generally, thisdetermination is based on the previous year’s actual or modeled air pollution concentration. Chayand Greenstone [6,7] and Henderson [16] document the effectiveness of nonattainment status inreducing total suspended particulates and ozone air pollution, respectively. The current studyempirically tests the effect of nonattainment status on SO2 concentrations relative to attainmentstatus.
2.2. Background on sulfur dioxide pollution
Sulfur dioxide is a colorless gas, which is odorless at low concentrations but pungent at very highlevels. It results naturally from sources such as volcanoes and decaying organic material. Humanactivity accounts for approximately 100 million tons of SO2 emissions per year. The primary man-made source of SO2 is from the combustion of fuels (e.g. coal and oil) that contain sulfur. Thelargest industrial emitters of SO2 are the electric utility plants that use cheaper high sulfur coal fromthe eastern US In the manufacturing sector, the largest emitters are petroleum refineries, smelters,paper mills, chemical plants, and producers of stone, clay, glass, and concrete products.SO2 has a number of unwanted effects. It is one of the major contributors to smog. At high
levels, it can affect the respiratory system (especially among asthmatics), weaken the lung’sdefenses, aggravate existing respiratory and cardiovascular diseases, and potentially even lead topremature death [23].5 At much lower levels, it damages trees and agricultural crops. Along withnitrous oxides, SO2 is the major contributor to the formation of acid rain.
3. Data and overview of trends in sulfur dioxide concentrations and regulation
To implement the analysis, I compiled the most detailed and comprehensive data availableon SO2 concentrations and nonattainment status. This section describes the data and its sources andpresents summary statistics on national trends in SO2 monitoring, concentrations, and regulation.
3.1. SO2 and regulation data
The SO2 concentration data were obtained by filing a Freedom of Information Act request withthe EPA. This yielded the Quick Look Report data file, which is derived from the EPA’s AirQuality Subsystem database. For each SO2 monitor operating at any point in the 1969–1997period, the file contains annual information on the number of recorded hourly readings, theaverage across all these readings, and the two highest readings for 1, 3, and 24 h periods. It alsoreports the monitor’s county.Annual county-level SO2 concentrations are calculated as the weighted average of the annual
arithmetic means of each monitor in the county, with the number of observations per monitorused as weights. In practice, I calculate this from the set of monitors that operated for at least
ARTICLE IN PRESS
5See Chay et al. [5], Chay and Greenstone [6,8], Dockery et al. [10], and Ransom and Pope III [19] on the relationship
between air pollution and human health.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611588
75% of the year (i.e., 6570 h). These monitors meet the EPA’s minimum summary criterion andthis sample restriction is applied in throughout the analysis.6
The annual county-level SO2 nonattainment designations were determined from the annualCode of Federal Regulations (CFR). For each of the criteria pollutants, the CFR lists every countyas, ‘‘does not meet primary standards’’, ‘‘does not meet secondary standards’’, ‘‘cannot beclassified’’, ‘‘better than national standards’’, or ‘‘cannot be classified or better than nationalstandards’’. Additionally, the CFR occasionally indicates that only part of a county did not meetthe primary standards. I assigned a county to the SO2 nonattainment category if all or part of itfailed to meet the SO2 ‘‘primary standards’’; otherwise, I designated it attainment. These annual,county-level, designations were hand entered for the 3063 US counties for the years 1978–1992.Although these designations were applied before 1978, the identity of the nonattainment countiesis unavailable from earlier CFRs and the EPA. This poses some difficulties for the analysis,because it means that there is no ‘‘pre-period’’ where the regulations are not in force. This issue isdiscussed further below.
3.2. National trends in SO2 monitoring, concentrations, and regulation
Table 1 presents annual summary information from the SO2 monitor and nonattainmentdata. Column (1) reports the number of monitors that meet the EPA’s minimum summarycriteria. Columns (2) and (3) present the number of counties in which these monitors are locatedand their population. The passage of the 1970 CAAAs led to an extensive program to sitemonitors around the US and obtain better readings of the prevailing SO2 concentrations.Since the late 1970s, the number of operating monitors has varied between approximately 500 and650. These monitors are located in roughly 300 counties, implying that on average there areslightly less than 2 monitors operating in each county.7 From 1980 onward, the population incounties monitored for SO2 ranges between 110 and 125 million, demonstrating that themonitoring program was focused on the most heavily populated of the more than 3000 UScounties.Although the total number of monitored counties is roughly constant after the late 1970s, there
is a lot of churning in the set of monitored counties. For example, 576 different counties aremonitored over the 1969–97 period. Since the monitoring program is aimed at the dirtiestcounties, the EPA and the states generally remove monitors from counties that become ‘‘clean’’and add monitors in ‘‘dirty’’ counties. In order to avoid any biases associated with compositionalchanges, the subsequent tests of the effect of nonattainment status on SO2 concentrations areconducted on fixed sets of counties. In the presence of the churning, however, this restrictionmeans that there is a tension between the number of years and counties in any sample.
ARTICLE IN PRESS
6To preclude the possibility that counties or states place monitors to produce false pollution concentrations, the Code
of Federal Regulations contains precise criteria that govern the siting of a monitor. Among the most important criteria is
that the monitors capture representative pollution concentrations in the most densely populated areas of a county.
Moreover, the EPA must approve the location of all monitors and requires documentation that the monitors are
actually placed in the approved locations. This information is derived from the Code of Federal Regulations 1995, title
40, part 58 and conversations with Manny Aquilania and Bob Palorino of the EPA’s District 9 Regional Office.7 In the vast majority of counties, only 1 monitor records SO2 concentrations. However, in large, industrial counties,
such as Los Angeles, CA, Cook (Chicago), IL, and Wayne (Detroit), MI, it is not uncommon for more than 10
monitors to operate simultaneously.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 589
ARTIC
LEIN
PRES
STable 1
Trends in EPA’s SO2 monitoring and regulatory programs, 1969–1997
Year Monitors Counties Population(millions)
SO2 concentrations (ppm) Number of counties exceeding standard for SO2 nonattain counties
Annualmean(unweighted)
Annualmean(populationweighted)
Meanof 2ndhighest 24 h
Year 24 h Year or 24 h Monitored Total
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
1969 43 16 23.4 0.0380 0.0359 0.1904 8 8 9 — —1970 30 17 22.6 0.0301 0.0266 0.1493 6 8 8 — —1971 73 44 34.5 0.0224 0.0210 0.1121 9 14 18 — —1972 107 57 44.9 0.0206 0.0190 0.1009 6 14 15 — —1973 144 87 53.1 0.0185 0.0172 0.0970 12 17 20 — —1974 271 146 70.2 0.0172 0.0149 0.0833 11 20 25 — —1975 319 168 68.3 0.0150 0.0136 0.0746 10 21 23 — —1976 425 204 88.0 0.0143 0.0131 0.0728 11 22 23 — —1977 447 205 84.4 0.0142 0.0121 0.0758 10 22 25 — —1978 494 225 90.3 0.0127 0.0114 0.0646 2 13 13 41 871979 559 252 94.3 0.0125 0.0110 0.0682 3 12 14 43 841980 561 274 110.3 0.0105 0.0098 0.0540 1 15 15 45 831981 593 287 114.7 0.0098 0.0093 0.0558 1 8 8 47 801982 588 299 113.2 0.0093 0.0085 0.0479 0 6 6 42 601983 629 331 117.1 0.0089 0.0083 0.0461 1 4 4 36 531984 640 329 116.9 0.0094 0.0087 0.0521 2 6 7 31 481985 577 305 114.6 0.0088 0.0085 0.0474 0 6 6 34 511986 562 309 120.4 0.0084 0.0083 0.0449 0 3 3 36 491987 523 299 120.6 0.0076 0.0080 0.0388 1 4 4 29 481988 528 309 124.5 0.0079 0.0079 0.0413 1 3 4 30 491989 533 309 119.9 0.0078 0.0079 0.0392 0 1 1 30 491990 523 310 122.5 0.0072 0.0074 0.0363 1 1 2 29 461991 535 315 125.6 0.0070 0.0073 0.0363 0 2 2 28 461992 522 308 123.4 0.0064 0.0066 0.0338 0 1 1 25 461993 524 320 125.5 0.0061 0.0064 0.0314 0 2 2 — —1994 516 309 120.3 0.0059 0.0062 0.0331 0 1 1 — —1995 537 320 126.2 0.0048 0.0050 0.0255 0 0 0 — —1996 532 320 127.8 0.0048 0.0050 0.0265 0 1 1 — —1997 511 310 122.1 0.0048 0.0050 0.0250 0 1 1 — —
Notes: The data sources are the EPA’s SO2 monitoring network and the Code of Federal Regulations (CFR). The EPA considers a monitor
‘‘representative’’ of the prevailing ambient SO2 concentration if it records at least 6570 (out of a maximum of 8760) hourly concentrations in a year.
The sample is restricted to monitors that meet this definition of representativeness in columns (1)–(9). Counties with at least 1 monitor satisfying this
sample restriction form the universe of counties in column (7), but data from all monitors within those counties is used to determine whether the 24 h
SO2 air quality standard is exceeded. Column (10) reports the number of nonattainment counties in the entire US. Nonattainment status was
published annually in the CFR beginning in 1978 and is unavailable before this year. The post-1992 data is currently unavailable in electronic form.
The 1970 Census’ population counts were used for the population measures from 1969 through 1975. The 1980 Census was used for 1976–1985 and
the 1990 Census for 1986–1997.
M.
Green
ston
e/
Jo
urn
al
of
En
viron
men
tal
Eco
no
mics
an
dM
an
ag
emen
t4
7(
20
04
)5
85
–6
11
590
Columns (4) and (5) report the unweighted and weighted (where county population is theweight) mean annual concentration across all operating monitors. Column (6) lists the annualmeans of the 2nd highest 24-h county-level readings. Each county’s 2nd highest reading is themaximum of the 2nd highest daily readings across all monitors (not just those that meet theminimum summary criteria) within a county by year and the entry is the unweighted mean ofthese readings.All three measures reveal a dramatic decline in SO2 concentrations in the last 30 years. For
example, the unweighted (population weighted) annual average declined from 0.0380 ppm(0.0359 ppm) in 1969 to 0.0105 (0.0098) in 1980 and then to 0.0048 (0.0050) in 1997. The extremeconcentration readings follow a similar pattern. It is evident that the national average SO2
concentration was well below the federal standards by the mid-1970s. Fig. 1 graphically displaysthe SO2 trends and the population in monitored counties. In light of the similarity in theunweighted and weighted concentrations, the subsequent analysis exclusively uses the unweightedmeasures.Columns (7)–(9) report the number of counties with monitor readings exceeding the annual,
daily, or either federal standard. All of these measures exhibit a strong downward trend. In factsince 1981, fewer than 10 of the monitored counties exceeded the SO2 NAAQS in any year. Andsince 1986, it is fewer than 5.Columns (10) and (11) list the number of counties designated nonattainment in the CFRs. The
former column reports this information among the set of monitored counties, while the lattercolumn reports it for all counties.8 The number of nonattainment counties always greatly exceeds
ARTICLE IN PRESS
0.045
0.040
0.035
0.030
0.025
0.020
0.015
0.010
0.005
0.000
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
Year
20
30
40
50
60
70
80
90
100
110
120
130
140
Population Weighted Arithmetic Mean Unweighted Arithmetic Mean Population in Monitored Counties
Anu
ual A
rith
met
ic M
ean
(ppm
)
Pop
ulat
ion
(Mil
lion
s)
Fig. 1. Trends in sulfur dioxide concentrations and population in monitored counties, 1969–1997.
8A comparison of the two columns reveals that many counties that do not meet the criterion for sufficient monitoring
are designated SO2 nonattainment. Presumably, the EPA’s models indicate that these counties have SO2 concentrations
that exceed the NAAQS.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 591
the number of counties exceeding the SO2 NAAQS in the previous year. In fact a number of thenonattainment counties had not exceeded the standard for many years.9
It is evident that SO2, nonattainment status is not mechanically assigned according to whether acounty violates the NAAQS. This finding greatly complicates an evaluation of the effect ofnonattainment status on SO2 concentrations. Since the NAAQS specify the selection rule that inprinciple determines nonattainment status, it is natural to evaluate this program with a regressiondiscontinuity approach as in [6,7]. Such an evaluation technique relies on the assumption thatcounties just above and below the NAAQS threshold are virtually identical and that comparisonsnear the threshold will produce unbiased estimates of the effect of nonattainment status on airpollution concentrations.This section’s findings reveal that the practical selection rule for the SO2 nonattainment
designation is unknown, so it is inappropriate to use a regression discontinuity design here.Consequently, the validity of the subsequent analysis turns on the successfulness of adjusting forthe heterogeneity across nonattainment and attainment counties. This is the subject of theremainder of the paper.
4. The research design and challenges to its validity
4.1. Research design
This paper conducts separate analyses of the impact of SO2 nonattainment status on SO2 airpollution in 3 nonoverlapping periods, 1975–1980, 1981–1986, and 1987–1992. In order to beincluded in the sample for a period, a county must have at least one monitor that meets theminimum summary criteria in every year of that period.10 The extension of the sample beyond 6years causes the sample sizes to deteriorate rapidly. Further, due to the small number ofmonitored counties in the early 1970s, it is not practical to begin the analysis until 1975, 5 yearsafter the passage of the 1970s CAAAs.The paper tests the effect of the nonattainment designation that is assigned in the 4th year of
each period. Since the nonattainment designation is supposed to be based on the previous year’spollution concentration the third year is henceforth referred to as the ‘‘regulation selection year’’or year ‘‘t21’’. Thus, each period contains 3 years before the assignment of this designation and 3years subsequent to it. The 3 ‘‘pre-period’’ years are used to adjust for differences across thenonattainment and attainment counties. The CAAAs frequently require the implementation ofabatement activities over a period of a few years and the 3 ‘‘post-period’’ years allow the effect tovary over time.
ARTICLE IN PRESS
9As an example, the paper examines 19 counties that are designated SO2 nonattainment in 1990. 12 of these 19
counties did not exceed the NAAQS in any of the previous 10 years.10The advantage of fixing the sample is that it avoids any biases associated with changes in the composition of
monitored counties. The disadvantage is that the EPA’s decision to stop monitoring a county may be associated with
nonattainment status. In particular, the nonattainment regulations may reduce SO2 concentrations and the EPA may
systematically cease monitoring the counties with these improvements in air quality. In this case, this paper’s use of a
fixed sample approach will cause the effect of the nonattainment designation to be biased upward.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611592
4.2. Pre-period differences in attainment and nonattainment counties
Since the nonattainment designation is not assigned randomly, this subsection examines theassociation between nonattainment status and the likely determinants of changes in SO2
concentrations in the post-period. Because the CAAAs require the assignment of nonattainmentstatus to be based on SO2 concentrations, it is to be expected that there are differences across thecategories of counties. The identification of these differences is important, because it reveals thesources of heterogeneity that must be adjusted for in order to avoid confounding the effect ofnonattainment status with them.Table 2 explores some of these sources of heterogeneity. Each 6-year period is examined in a
separate panel. Columns (1) and (2) detail the number of counties and their populations in eachperiod. Columns (3a)–(3e) report the mean SO2 concentration in the regulation selection year (i.e.,1977, 1983, and 1989), the change between t21 and t23; and the number of counties exceedingthe SO2 NAAQS in each of the pre-period years, respectively.Figs. 2 and 3 graphically describe the pre-period heterogeneity. Fig. 2 plots the annual SO2
concentrations by nonattainment status. The points to the right of the vertical line are the post-period years. Fig. 3 presents separate kernel density plots of the distribution of SO2
concentrations in the regulation selection year for nonattainment and attainment counties. Bothfigures include separate panels for each period.
1975–80: This period is structured to take advantage of the first publication of thenonattainment designations in 1978. It may be an especially interesting period, because it allowsfor a test of nonattainment status in the first year that the 1977 CAAAs were in force. There are 44attainment and 18 nonattainment counties, with populations of 23.6 and 15.7 million,respectively. A simple before and after comparison suggests that nonattainment status reducedSO2 concentrations.There are at least two important differences in the pre-period concentrations of SO2. First,
mean SO2 concentrations are roughly two times higher in nonattainment counties. From thekernel density plots in Fig. 3A, it is evident that this mean difference is not due to a few outliers. Inparticular, there are few attainment counties with concentrations in the range covered bythe majority of the nonattainment distribution. Second, there is a downward trend in thenonattainment counties in the pre-period, while concentrations are essentially flat in theattainment ones.Columns (3c)–(3e) of Table 2 reveal that only 7 of the 18 nonattainment counties meet the
statutory requirements necessary for this designation. Among the attainment counties, only 1 ofthem exceeds the federal standards. Overall, it is evident that the practical selection rule fornonattainment status differed from the statutory one.In summary, Fig. 2A suggests that 1978 nonattainment status reduced SO2 concentrations, but
a closer examination of the data leads to a less certain conclusion. In particular, it is possible thatthe observable differences among attainment and nonattainment counties or the unobservedvariables that determine the selection rule for nonattainment status explain the relative post-period decline in nonattainment counties. It is evident that in order to obtain a consistent estimateof nonattainment status, the subsequent analysis will need to successfully adjust for these factors.
1981–86: The 1980s are often regarded as a period when environmental regulations were notenforced vigorously. This period provides an opportunity to empirically test this hypothesis.
ARTICLE IN PRESS
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 593
ARTIC
LEIN
PRES
S
Table 2
Pre-period characteristics of counties, by nonattainment status
Counties Population Pre-period SO2 concentration summary statistics SO2 nonattainment summary statistics
Year t-1 Mean
Year
t-1
Pre-period
change
(year
t-1–year t-3)
Exceed annual or
24 h SO2 NAAQS
At least 1
pre-period
year
All pre-period
years
Every year 1978
through year
t-1
Year
t-1
Year
t-2
Year
t-3
(1) (2) (3a) (3b) (3c) (3d) (3e) (4a) (4b) (4c)
1975–1980
Nonattainment 18 15,728,538 0.0220 �0.00268 7 6 7 — — —
Attainment 44 23,567,800 0.0117 0.00053 1 2 2 — — —
1981–1986
Nonattainment 23 9,709,513 0.0110 �0.00369 1 3 2 23 23 20
Attainment 139 76,277,154 0.0091 �0.00083 1 2 3 10 2 2
1987–1992
Nonattainment 19 7,719,144 0.0125 �0.00196 1 2 3 19 17 13
Attainment 184 87,390,332 0.0084 0.00008 1 2 1 5 1 1
Note: Each panel reports summary information from the samples for each of the three 6 year periods analyzed in the paper. Column (1) reports the
number of counties in the attainment and nonattainment categories in 1978, 1984, and 1990, respectively. Column (2) reports the population in each
set of counties. Columns (3a)–(3e) report summary information on SO2 concentrations. Year t-1 refers to the regulation selection year, which is the
year before the assignment of the nonattainment designation (i.e., 1977, 1983, and 1989). Years t-2 and t-3 are 2 and 3 years before the regulation
selection year, respectively. Columns (4a)–(4c) report the number of counties designated nonattainment in any of the pre-period years (i.e., t-1; t-2;and t-3), each of the pre-period years, and in all years from 1978 through year t-1; respectively.
M.
Green
ston
e/
Jo
urn
al
of
En
viron
men
tal
Eco
no
mics
an
dM
an
ag
emen
t4
7(
20
04
)5
85
–6
11
594
ARTICLE IN PRESS
0.005
0.01
0.015
0.02
0.025
0.03
1975 1976 1977 1978 1979 1980
Mea
n A
nnua
l Sul
fur
Dio
xide
Con
cent
ratio
n (p
pm)
Attainment (44 Counties, 23.6 million) Nonattainment (18 Counties, 15.7 million)
Year
1978 Nonattainment Status (A)
1984 Nonattainment Status
0.007
0.008
0.009
0.01
0.011
0.012
0.013
0.014
0.015
0.016
1981 1982 1983 1984 1985 1986
Year
Mea
n A
nnua
l Sul
fur
Dio
xide
Con
cent
rati
on (
ppm
)
Attainment (139 Counties, 76.3 million) Nonattainment (23 Counties, 9.7 million)
(B)
0.007
0.008
0.009
0.01
0.011
0.012
0.013
0.014
0.015
0.016
Year
Mea
n A
nnua
l Sul
fur
Dio
xide
Con
cent
rati
on (
ppm
)
Attainment (184 Counties, 87.4 million) Nonattainment (19 Counties, 7.7 million)
(C)
1987 1988 1989 1990 1991 1992
1990 Nonattainment Status
Fig. 2. Trends in annual mean sulfur dioxide concentrations (ppm), by nonattainment status. (A) 1978 nonattainment
status, (B) 1984 nonattainment status, (C) 1990 nonattainment status.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 595
ARTICLE IN PRESS
1983 SO2 Concentrations by 1984 Nonattainment Status
1983 SO2 Concentration (ppm)
1984 SO2 Attainment Density 1984 SO2 Nonattainment Density
0 .01 .02 .03 .04 .05 .06 .07
0
20
40
60
80
(B)
1977 SO2 Concentrations by 1978 Nonattainment Status
1977 SO2 Concentration (ppm)
1978 SO2 Attainment Density 1978 SO2 Nonattainment Density
0 .01 .02 .03 .04 .05 .06 .07
0
20
40
60
80
(A)
1989 SO2 Concentrations by 1990 Nonattainment Status (C)
1989 SO2 Concentration (ppm)
1990 SO2 Attainment Density 1990 SO2 Nonattainment Density
0 .01 .02 .03 .04 .05 .06 .07
0
50
100
Fig. 3. Kernel density plots of the distribution of SO2 concentrations in the regulation selection years. (A) 1977 SO2
concentrations by 1978 nonattainment status, (B) 1983 SO2 concentrations by 1984 nonattainment status, (C) 1989 SO2
concentrations by 1990 nonattainment status.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611596
There are 139 counties with 76.3 million people in the 1984 attainment category and 23 countieswith a combined population of 9.7 million in the nonattainment one. In the years in which1984 nonattainment status applied, the nonattainment counties have a slightly larger declinein SO2.Again, there are important pre-period characteristics of the counties that vary with
nonattainment status. For example, the regulation selection year mean SO2 concentrations arehigher in nonattainment counties. Also, the pre-period decline is greater in the nonattainmentcounties (roughly 0.0037 vs. 0.0008 ppm). Notably, only 1 of the nonattainment counties exceededthe SO2 NAAQS in the regulation selection year. This finding is salient because it indicates thatthe statutory selection rule did not determine the assignment of SO2 nonattainment in this periodtoo. This reinforces the point that a regression discontinuity evaluation strategy is notappropriate.
1987–92: This period allows for a test of the effect of nonattainment status in the first years inwhich the 1990 CAAAs were in force. There are 184 counties with a population of 87.4 million inthe 1990 attainment category and 19 counties with approximately 7.7 million people in thenonattainment group. Together these counties accounted for approximately 40% of the USpopulation in this period. Fig. 2C reveals an association between nonattainment status and a post-period reduction in SO2 concentrations.However, there are substantial pre-period differences in the two sets of counties. In the
nonattainment counties during the pre-period, SO2 concentrations are higher and there is a largerdownward trend. Further, the rule that determines nonattainment status remains a mystery. Itappears likely that just as in the other periods, a simple unadjusted pre- and post-comparison willnot reveal the causal effect of nonattainment status on SO2 concentrations.
4.3. Intra-county variation in nonattainment status over time
An important issue for this analysis is that there is little intra-county variation in counties’attainment–nonattainment designation over time. This is because few counties’ nonattainmentstatus changed after 1978. The last three columns of Table 2 demonstrate this point in the threeperiods examined here. Columns (4a), (4b), and (4c) report the number of counties designatednonattainment in any of the pre-period years (i.e., t21; t22; and t23), all the pre-period years,and in all years from 1978 through year t21; respectively. Column (4b) reveals that almost allcounties that are nonattainment in the 4th year of the 1981–6 and 1987–92 periods are alsononattainment throughout the pre-period. Further, 20 (13) of 23 (19) 1984 (1990) SO2
nonattainment counties were also nonattainment in every year from 1978 through the regulationselection year. Interestingly, a small, but nontrivial, fraction of the attainment counties werepreviously designated nonattainment.The absence of time variation in nonattainment status has a number of important implications.
First, the ‘‘pre-period’’ years are not before the counties are regulated, but simply before thearbitrarily chosen 1978, 1984, and 1990 nonattainment designations that this paper evaluates. Forease of exposition, I use ‘‘pre’’ and ‘‘post’’ throughout the remainder of the paper but note thattheir meanings here differs from the standard definitions.Second, although it would be interesting to determine whether there is heterogeneity in the
effect of nonattainment status, it will be difficult to investigate some likely sources. For example,
ARTICLE IN PRESS
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 597
the effect might vary with the number of years that a county has been designated nonattainment(i.e., the effect on SO2 concentrations may be greater in the 1st year that a county isnonattainment than in the 9th year). Alternatively, the EPA’s enforcement intensity may varywith the political climate. Unfortunately, it is impossible to separately identify these two potentialsources of heterogeneity because almost all counties that were ever nonattainment were firstdesignated nonattainment in 1978. The subsequent analysis allows the effect of nonattainmentstatus to differ in each of the three periods but whether any differences are due to the length oftime that a county was designated nonattainment or presidential preferences about theenvironment cannot be assessed.Third, and most importantly, the analysis is structured to evaluate the effect of the
nonattainment designation assigned in the 4th year of each period, but this interpretationof the estimated effects may not be valid. Although linear regression and propensity scorematching are used to adjust for the pre-period differences (e.g., in SO2 concentrations) betweenthe nonattainment and attainment counties, the absence of within county variation innonattainment status means that it is impossible to directly control for pre-period differencesin the incidence of the nonattainment designation. Consequently, it may be more appropriate tointerpret the estimated treatment effect as a measure of the effect of a county being designatednonattainment over a longer period, rather than the effect of the 4th year’s nonattainmentdesignation.
5. Econometric strategies
This paper utilizes a variety of statistical models to control for the differences betweenattainment and nonattainment counties that were documented in the previous section. Here, Idescribe these econometric strategies.
5.1. First differences
The first differences strategy is based on the assumption that any differences in SO2
concentrations between the two sets of counties are due to permanent factors (e.g., a county’stopography) that have time invariant effects on SO2 pollution. Let Di4 be an indicator variablethat equals 1 when county i is designated nonattainment in the 4th year of one of the 6 yearperiods and 0 if it is attainment. Now, consider:
Yit � Yi3 ¼ yt þ atDi4 þ Zit; ð1Þ
where t denotes the year and 4ptp6; indicating that the dependent variable is the change in SO2
concentrations between the 4th, 5th, or 6th year and the 3rd year. yt is a nonparametric time effectthat is common to attainment and nonattainment counties and Zit is the idiosyncratic unobservedcomponent of the change in SO2 concentrations.at is the parameter of interest and measures the difference in the change in SO2 concentrations
between nonattainment and attainment counties. Formally, at ¼ E½Yit � Yi3jDi4 ¼ 1 � E½Yit �Yi3jDi4 ¼ 0: Eq. (1) is estimated separately for each of the post-period years and the t subscripton a highlights that the effect of nonattainment status is allowed to vary across years within a
ARTICLE IN PRESS
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611598
period.11 For example, it may be of a larger magnitude in the 6th year of a period (e.g., 1980) thanin the 4th (e.g., 1978). Notably, the differencing removes the time-invariant characteristics that arethe only source of bias under this model’s assumptions.The mapping that assigns a value to the nonattainment indicator merits further discussion. This
mapping emphasizes that the aim of this exercise is to evaluate the effect of the 4th yearnonattainment designation. The difficulty is that the pre-period SO2 nonattainment designationsare highly correlated with Di4: This correlation will cause the estimated at to capture the effect ofthe 4th year nonattainment designation and the effect of the pre-period SO2 nonattainmentdesignations. The point is that it is likely to be incorrect to interpret at as the effect of the 4th yearnonattainment designation only.A more troubling possibility is that the identifying assumptions of this model are unlikely to be
valid. This is because the nonattainment designation is highly correlated with past pollution levels,which in turn are likely correlated with current pollution levels through Zit: This would be the caseif, for example, there is mean reversion in SO2 concentrations. Moreover, the differential pre-period trends in SO2 concentrations documented in Section 4 are likely to cause a correlationbetween Di4 and Zit: Such unwanted correlations will bias the estimated effect of nonattainmentstatus, at:
5.2. Adjusted first differences
In light of these issues, the paper also linearly adjusts for the heterogeneity across attainmentand nonattainment counties. The model is
Yit � Yi3 ¼ yt þ X 0ibbþ P0
ibjþ atDi4 þ Zit; ð2Þ
where the b subscript indicates the pre-period. The substantive difference with Eq. (1) is that herethe estimated effect of nonattainment status on the change in SO2 concentrations is conditional onlinear adjustment for the vectors Xib and Pib:In practice, Xib includes nonattainment status for carbon monoxide, ozone, and total suspended
particulates in the same year that Di4 is measured (i.e., 1978, 1984, and 1990). As Table 2demonstrated, there is little within county variation in SO2 nonattainment status, so it is notpractical to include measures of this variable from earlier years. The vector also includes county-level measures of per capita income, the average wage, total employment, and total population.12
Moreover in some specifications, Xib includes a full set of state fixed effects so that the estimatedtreatment effect is based on intra-state comparisons of SO2 attainment and nonattainmentcounties.
Pib is a vector that includes a number of measures of lagged SO2 pollution concentrations inorder to break the unwanted correlation between Di4 and Zit: For instance, it includes the meanSO2 concentration from the t22 and t23 pre-period years (e.g., 1976 and 1975 in the 1975–80period). The vector also includes the annual maximum 1, 3, and 24 h readings from t21; the
ARTICLE IN PRESS
11For example in the 1975–80 period, I separately estimate (1) when the dependent variables are Yi1978 � Yi1977;Yi1979 � Yi1977; and Yi1980 � Yi1977:
12These data are obtained from the Bureau of Economic Analysis.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 599
regulation selection year. Since the correct functional form for the lagged SO2 pollution variablesis unknown, some specifications will include squares and cubes of these variables.Here, the identifying assumption is that E½Di4ZitjXib;Pib; yt ¼ 0 or that 4th year nonattainment
status is independent of potential pollution concentrations, conditional on the covariates. Recall,the EPA’s exact selection rule is unknown but it seems sensible to assume that it is a function oflagged pollution levels, economic activity, and its cognizance of a county’s overall air quality(measured by the nonattainment designations for other pollutants). If the EPA relies on othervariables that also determine post-period SO2 concentrations, the assumption will be violated andthe estimated treatment effect will be inconsistent.
5.3. Propensity score matching
Eq. (2) relies on a linear model to control for the covariates Xib and Pib: This may beunappealing since their true functional form is unknown. An alternative is to compare the changein SO2 concentrations of attainment and nonattainment counties that have identical values of Xib
and Pib: This would obviate all functional form concerns. Of course, this method is not feasiblewhen there are many variables or even a few continuous variables, because it becomes impossibleto match nonattainment and attainment counties. This problem is known as the ‘‘curse ofdimensionality’’.As a solution, Rosenbaum and Rubin [20] suggest matching on the propensity score—the
probability of receiving the treatment conditional on covariates.13 This probability is an index ofall covariates and effectively compresses the multi-dimensional vector of covariates into a simplescalar. The advantage of the propensity score approach is that it offers the promise of providing afeasible method to control for the observables in a more flexible manner than is possible withlinear regression. Just as with linear regression, the identifying assumption is that assignment tothe treatment is associated only with observable pre-period variables. This is often called theignorable treatment assignment assumption or selection on observables [15].I implement the propensity score matching strategy in three steps. First, the propensity score is
obtained by fitting logits for SO2 nonattainment status, using Xib and Pib as explanatory variables.Thus, I try to replicate the EPA’s selection rule with the observed covariates. I then conduct twotests. The first examines whether the estimated propensity scores of the nonattainment andattainment counties are equal quintiles. The second tests whether the means of the covariates areequal in the nonattainment and attainment counties within quintiles. If the null hypothesis is notaccepted for either test, I divide the quintiles and/or estimate a richer logit model by includinghigher order terms and interactions.14 Once the null is accepted for both tests, I proceed to thenext step.Second, the treatment effect is calculated by comparing the change in SO2 concentrations
between a post-period year and year t21 of nonattainment and attainment counties with similaror ‘‘matched’’ values of the propensity score. In particular, a treatment effect is calculated for each
ARTICLE IN PRESS
13An alternative is to match on a single (or possibly a few) ‘‘key’’ covariate(s). See Angrist and Lavy [2] or Rubin [22]
for applications.14See Dehejia and Wahba [9] and Rosenbaum and Rubin [21] for more details on how to implement the propensity
score method.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611600
nonattainment county for which there is at least one attainment county with a propensity score‘‘suitably close’’ to the value for the nonattainment county. In the subsequent results, I definesuitably close in two ways—within calipers of 0.05 and 0.10. In cases where multiple attainmentcounties fall into one of these calipers, I average the outcome across all of these counties. Further,this matching is done with replacement so that individual attainment counties can be used ascontrols for multiple nonattainment counties.15
Third, a single treatment effect is estimated by averaging the treatment effects across all thenonattainment counties for which there was at least one suitable match. An attractive feature ofthis approach is that the estimated treatment effect is based on comparisons of nonattainment andattainment counties with ‘‘similar’’ values of the index. This has the desirable property that itfocuses the comparisons where there is overlap in the distribution of the propensity scores in thenonattainment and attainment samples.16
6. Empirical results
6.1. 1975–1980
Here, I present evidence on the relationship between 1978 SO2 nonattainment status and SO2
concentrations in 1978, 1979 and 1980. Table 3A contains the linear regression results. The entriesare the parameter estimate from the 1978 nonattainment indicator, its estimated standard error
(in parentheses), and the R2 statistic. Each set of entries is from a separate regression. In the first,second, and third panels, the dependent variables are the 1978–77, 1979–77, 1980–77 change inmean SO2 concentrations.Column (1) reports the unadjusted first difference estimate from the fitting of Eq. (1). These
estimates indicate that SO2 concentrations declined by 0.00259, 0.00290, and 0.00477 ppm more innonattainment counties than attainment counties in 1978, 1979, and 1980 (relative to theregulation selection year), respectively. These results suggest that by the end of the period (i.e.,1980) nonattainment status is associated with a 17% decline in SO2 concentrations. Theimportant limitation of the first difference estimator is that it does not adjust for time-varyingfactors (e.g., the pre-existing trends in the two sets of counties). As Table 2 and Fig. 2Ademonstrated, the pre-period trends in these two sets of counties are starkly different.The entries in the remaining columns are from the estimation of Eq. (2) and they progressively
control for more potential confounders. In the column (2) specification the lagged measures ofSO2 are included, while in column (3) the measures of nonattainment status for the otherpollutants are added. The column (4) specification includes the economic activity controls andlastly column (5) adds a full set of state fixed effects. The exact controls are noted at the bottom ofthe table.17
ARTICLE IN PRESS
15See Dehajia and Wahba [9] and Heckman, Ichimura, and Todd [13] on propensity score matching algorithms.16 If there are heterogeneous treatment effects, this strategy produces an estimate of the average ‘‘effect of the
treatment on the treated’’ when there are suitable matches for all nonattainment counties [1,4,14,22].17 In this period the lagged pollution variables enter linearly but higher order terms are not included due to the small
sample size.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 601
ARTICLE IN PRESS
Table 3
Estimated effect of 1978 nonattainment status on mean SO2 concentrations
(A) First differences estimates
(1) (2) (3) (4) (5)
(1978 SO2–1977 SO2)/1001978 nonattainment �0.259 0.077 0.100 0.121 0.083
(0.138) (0.101) (0.096) (0.097) (0.098)
R2 0.09 0.66 0.71 0.73 0.87
(1979 SO2–1977 SO2)/1001978 nonattainment �0.290 0.037 0.095 0.139 0.154
(0.181) (0.131) (0.142) (0.144) (0.175)
R2 0.06 0.62 0.64 0.67 0.78
(1980 SO2–1977 SO2)/1001978 Nonattainment �0.477 �0.082 0.078 0.089 0.100
(0.201) (0.153) (0.159) (0.173) (0.213)
R2 0.12 0.55 0.61 0.64 0.81
N 62 62 62 62 62Lagged SO2 concentrations No yes Yes Yes YesFlexible functional forms of laggedSO2 concentrations
No no No No No
Nonattainment for other pollutants No no Yes Yes YesEconomic activity controls No no No Yes YesState fixed effects No no No No Yes
Notes: The entries are the parameter estimates from the 1978 nonattainment indicator variable and the associated standard errors fromseparate regressions. The standard errors are calculated with the Eicker–White formula to account for unspecified heteroskedasticity.See the text, particularly the description of Eqs. (1) and (2), for further details. The included covariates are reported at the bottom ofthe table.
(B) Propensity score matching estimates
Logit Specification 1 Logit Specification 2
Caliper=0.10 Caliper=0.05 Caliper=0.10 Caliper=0.05(1) (2) (1) (2)
(1977 SO2)/1001978 nonattainment 0.402 0.538 0.314 0.113
(0.353) (0.426) (0.463) (0.644)
(1978 SO2–1977 SO2)/1001978 nonattainment 0.026 �0.063 �0.264 �0.607
(0.127) (0.139) (0.157) (0.109)
(1979 SO2–1977 SO2)/1001978 nonattainment �0.070 �0.102 �0.213 �0.711
(0.183) (0.183) (0.239) (0.209)
(1980 SO2–1977 SO2)/1001978 nonattainment �0.252 �0.331 �0.289 �0.879
(0.231) (0.246) (0.357) (0.482)
# of 18 treated that are matched 11 8 6 3# of 44 controls that are matched 44 43 4 2
Notes: The entries are the treatment effects and their standard errors from the propensity score matching procedure described in thetext. Each nonattainment county is matched to all the attainment counties with propensity scores within the specified caliper. LogitSpecification 1 includes the controls from the column (2) specification of panel (A) of this table; plus the mean SO2 concentration in theregulation selection year. Logit Specification 2 includes the controls from the column (4) specification of panel (A), plus the mean SO2
concentration in the regulation selection year.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611602
The key finding from columns (2) to (5) is that, once the lagged pollution levels are included ascovariates, the nonattainment designation is associated with modest increases in SO2. However,
none of these estimates would be judged statistically significant at conventional levels.18 The R2
statistics highlight that the addition of these variables greatly improves the fit of the regressions.Table 6 in the appendix reports the coefficients on the control variables from the fitting of thecolumn (4) specification when the dependent variable is the difference in the 1980 and 1977 SO2
concentrations.Table 3B presents the results from the propensity score matching routine. This is done with two
different specifications of the logit. In ‘‘Logit Specification 1’’, the covariates in the logit are thelagged SO2 measures (including the 1977 mean SO2 concentration), while ‘‘Logit Specification 2’’adds the county level nonattainment designations for the other criteria pollutants and theeconomic activity controls. The set of covariates in the logits correspond to those in columns (2)and (4) of Table 3A, respectively. For each specification of the logit, the Table reports theestimated treatment effects and their standard errors based on calipers of 0.10 and 0.05.19 Thebottom two rows report the number of nonattainment counties for which at least one match isavailable and the number of attainment counties that fall within the calipers, respectively.The first panel reports the results from a test of whether SO2 concentrations in the regulation
selection year are equal among the matched comparisons. These results are presented, because thisis the most important difference between the two groups before the assignment of thenonattainment designation. They indicate that even after matching on the propensity score, the1977 SO2 concentrations are higher in the nonattainment counties, although these differenceswould not be judged statistically different from zero at conventional levels. The differences rangebetween 0.0011 and 0.0054 ppm, depending on the specification of the logit and the size of thecaliper. Notably, these differences are substantially smaller than the average difference of0.0103 ppm across attainment and nonattainment counties in the complete sample. Thus, thematching has been relatively successful on this dimension.Before proceeding to the main matching results, Fig. 4A displays the overlap in the
distributions of the estimated propensity scores of the nonattainment and attainment countiesbased on Logit Specification 2. The figure contains the fraction and number of counties in eachquintile, as well as the means of the propensity score by attainment status. The overlap is less thanideal here. For example, the highest quintile is entirely comprised of nonattainment counties andthe mean propensity score among these counties is 0.97. Further, 39 of the 44 attainment countiesare in the lowest quintile. It is evident that it will not be possible to find suitable matches for manyof the nonattainment counties.Returning to Table 3B, the remaining panels present the results from tests of whether
nonattainment status is associated with relative declines in SO2 concentrations in the post-treatment years. Due to the difference in the regulation selection year SO2 concentrations across
ARTICLE IN PRESS
18 In all three periods, the results are qualitatively unchanged when the regressions are weighted by the county
population.19The standard errors are calculated according to the standard two-sample formula. These are generally considered
to be conservative estimates of the standard error for the estimated treatment effect, because they treat the matched
treatments and controls as being independent of each other. However, the matching process likely creates some positive
dependence between them and this dependence is not accounted for.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 603
the matched counties, the outcome is equal to the post-treatment value of SO2 minus the value inthe regulation selection year. Again, separate results are reported for 1978, 1979, and 1980.The results differ dramatically from the linear regression approach. They suggest that
nonattainment status is associated with declines in SO2 pollution in this period. Logit
ARTICLE IN PRESS
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
Frac
tion
of C
ount
ies
in R
egul
ator
y C
ateg
ory
Quintile [Mean of Predicted Probability for Attainment Counties (Number of Counties); Nonattainment Counties]
I [0.02 (39); 0.17 (1)] II [0.23 (1); --- (0)] III [0.47 (1); 0.48 (4)]IV [0.74 (3); 0.65 (1)] V [--- (0); 0.97 (12)]
1978 Nonattainment Status
(A)
Frac
tion
of C
ount
ies
in R
egul
ator
y C
ateg
ory
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
Quintile [Mean of Predicted Probability for Attainment Counties (Number of Counties); Nonattainment Counties]
1984 Nonattainment Status
I [0.02 (124); 0.08 (2)] II [0.29 (7); 0.34 (4)] III [0.55 (3); 0.49 (3)] IV [0.66 (1); 0.67 (3)] V [0.80 (1); 0.93 (11)]
(B)
Frac
tion
of C
ount
ies
in R
egul
ator
y C
ateg
ory
Attainment Nonattainment
1990 Nonattainment Status
Quintile [Mean of Predicted Probability for Attainment Counties (Number of Counties); Nonattainment Counties]
(C)
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
I [0.04 (171); 0.07 (6)] II [0.27 (5); 0.32 (3)] III [0.51 (5); 0.52 (5)] IV [--- (0); --- (0)] V [--- (0); 0.98 (5)]
Fig. 4. Distribution of predicted probability of SO2 nonattainment status, by observed SO2 nonattainment status. (A)
1978 nonattainment status, (B) 1984 nonattainment status, (C) 1990 nonattainment status. Notes: The predicted
probabilities are based on Logit Specification 2. See the text for more details.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611604
Specification 1 indicates that by 1980, SO2 concentrations in 1978 SO2 nonattainment countiesdeclined by roughly 0.003 ppm, while the specification 2 estimates indicate a decline that rangesfrom 0.003 to 0.009 ppm. However, conventional criteria imply that all of the estimated declinesare indistinguishable from zero in the 3rd post-period year. And, the small number of matchedcounties in Logit 2 limits the generality of the findings. Nevertheless, the contrast between theseestimated treatment effects and those from linear regression is striking and underscores the poorreliability of linear regression when there is little overlap in the distributions of the covariates.20
6.2. 1981–1986
Tables 4A and B and Fig. 4B are structured identically to Table 3A and B and Fig. 4A. Theonly difference with the analysis of the previous period is that the sample sizes are large enough toadjust for the lagged SO2 concentrations with a more flexible functional form. In particular, theadjusted first difference estimates (in columns 2–5) and both of the logit specifications includequadratics and cubics of the lagged mean SO2 concentrations and quadratics of the extremevalues measures. Otherwise, the specifications are identical.The results based on linear adjustment imply that nonattainment status is associated with
modest reductions in SO2 concentrations. Almost the entire post-period decline has occurred inthe first post-treatment year. Interestingly, the estimates are roughly constant in columns (2)–(4).21 The estimates are also insensitive to the inclusion of a full set of state fixed effects as incolumn (5). Overall, the linear regression results suggest that nonattainment status is associatedwith a 6–9% decline in SO2 concentrations by the end of the period. Again, none of the estimatedparameters are statistically significant.A few points emerge from the propensity score matching procedure. First, the matching almost
completely eliminates the differences in regulation selection year SO2 concentrations. Second,there is more overlap in the distribution of the propensity scores than in the 1975–80 period butthere are still substantial differences in the two distributions. For example, 11 of the 12 counties inthe highest quintile are nonattainment. Third, the treatment effects are smaller than the ones fromlinear regression and imply that nonattainment status had little effect on SO2 concentrations inthis period.
6.3. 1987–1992
The 1987–92 period results are presented in Tables 5A and B and Fig. 4C. The estimates basedon linear adjustment suggest that nonattainment status is associated with a 0.0009–0.0014 ppmreduction in SO2 pollution by the end of the period. These estimates would generally beconsidered statistically different from zero. Interestingly, the largest estimate comes from the
ARTICLE IN PRESS
20There are at least two explanations for the differences between the estimated treatment effects from the linear
regression and propensity score approaches. First, the propensity score approach nonparametrically controls for
observable variables while the linear regression is necessarily parametric. Second, the matching estimate is not based on
the full set of nonattainment counties due to the poor overlap in the distributions of the propensity scores.21The parameter estimates and standard errors from the other covariates in the column (4) specification when the
dependent variable is the difference in the 1986 and 1983 SO2 concentrations are reported in Appendix Table 1. This
table also reports this information for the analogous regression from the 1987–1992 period.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 605
ARTICLE IN PRESS
Table 4
Estimated effect of 1984 nonattainment status on mean SO2 concentrations
(A) First differences estimates
(1) (2) (3) (4) (5)
(1984 SO2–1983 SO2)/100
1984 nonattainment 0.010 �0.054 �0.065 �0.058 �0.061(0.046) (0.056) (0.056) (0.058) (0.069)
R2 0.00 0.12 0.15 0.17 0.63
(1985 SO2–1983 SO2)/100
1984 nonattainment 0.077 �0.017 �0.029 �0.026 �0.032(0.061) (0.066) (0.068) (0.071) (0.071)
R2 0.01 0.22 0.24 0.25 0.75
(1986 SO2–1983 SO2)/100
1984 nonattainment �0.025 �0.098 �0.078 �0.085 �0.069(0.053) (0.057) (0.060) (0.061) (0.070)
R2 0.00 0.27 0.28 0.29 0.65
N 162 162 162 159 159
Lagged SO2 Concentrations No Yes Yes Yes Yes
Flexible functional forms of lagged
SO2 concentrations
No Yes Yes Yes Yes
Nonattainment for other pollutants No No Yes Yes Yes
Economic activity controls No No No Yes Yes
State fixed effects No No No No Yes
(B) Propensity score matching estimates
Logit Specification 1 Logit Specification 2
Caliper=0.10 Caliper=0.05 Caliper=0.10 Caliper=0.05
(1) (2) (1) (2)
(1983 SO2)/100
1984 nonattainment 0.100 0.067 0.071 0.087
(0.167) (0.167) (0.162) (0.181)
(1984 SO2–1983 SO2)/100
1984 nonattainment �0.006 0.003 �0.013 0.003
(0.072) (0.074) (0.072) (0.076)
(1985 SO2–1983 SO2)/100
1984 nonattainment 0.041 0.046 0.028 0.082
(0.086) (0.087) (0.084) (0.081)
(1986 SO2–1983 SO2)/100
1984 nonattainment �0.022 �0.023 �0.038 0.002
(0.085) (0.086) (0.082) (0.088)
# of 23 treated that are matched 16 15 16 14
# of 139 controls that are matched 139 116 136 124
Notes: See the notes to Tables 3A and B and the text.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611606
ARTICLE IN PRESS
Table 5
Estimated effect of 1990 nonattainment status on mean SO2 concentrations
(A) First differences estimates
(1) (2) (3) (4) (5)
(1990 SO2–1989 SO2)/100
1990 nonattainment 0.024 �0.053 �0.054 �0.050 �0.065(0.054) (0.035) (0.037) (0.037) (0.056)
R2 0.00 0.28 0.29 0.30 0.58
(1991 SO2–1989 SO2)/100
1990 nonattainment �0.077 �0.120 �0.122 �0.132 �0.183(0.038) (0.041) (0.043) (0.043) (0.058)
R2 0.02 0.39 0.39 0.40 0.68
(1992 SO2–1989 SO2)/100
1990 nonattainment �0.136 �0.090 �0.089 �0.106 �0.144(0.070) (0.041) (0.042) (0.043) (0.053)
R2 0.03 0.54 0.54 0.56 0.80
N 203 203 203 200 200
Lagged SO2 Concentrations No Yes Yes Yes Yes
Flexible functional forms of lagged
SO2 concentrations
No Yes Yes Yes Yes
Nonattainment for other pollutants No No Yes Yes Yes
Economic activity controls No No No Yes Yes
State fixed effects No No No No Yes
(B) Propensity score matching estimates
Logit Specification 1 Logit Specification 2
Caliper=0.10 Caliper=0.05 Caliper=0.10 Caliper=0.05
(1) (2) (1) (2)
(1989 SO2)1/100
1990 nonattainment 0.050 0.013 0.119 0.121
(0.151) (0.152) (0.162) (0.160)
(1990 SO2–1989 SO2)1/100
1990 nonattainment 0.001 0.001 0.001 �0.002(0.042) (0.042) (0.046) (0.046)
(1991 SO2–1989 SO2)1/100
1990 nonattainment �0.034 �0.033 �0.038 �0.036(0.056) (0.055) (0.061) (0.062)
(1992 SO2–1989 SO2)1/100
1990 nonattainment �0.053 �0.047 �0.065 �0.067(0.064) (0.064) (0.070) (0.071)
# of 19 treated that are matched 15 15 14 14
# of 184 controls that are matched 184 183 181 181
Notes: See the notes to Tables 3A and B and the text.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 607
specification that includes state fixed effects when the estimate is identified from comparisonsamong nonattainment and attainment counties within the same state. These estimates suggest that1990 nonattainment status reduced SO2 concentrations by 7–11% in nonattainment counties by1992.Fig. 4C demonstrates that the overlap of the distributions is better in this period than the other
two. Although there are not any attainment counties in the two highest quintiles, 14 of the 19nonattainment counties are in the lowest three quintiles where the entire attainment distribution islocated. The estimated treatment effects are about half the magnitude of the linear regressiontreatment effects and are not statistically distinguishable from zero. This is likely driven by theinability to find suitable comparisons for the nonattainment counties with the highest probabilityof being designated nonattainment. Overall, the estimates from this period suggest that 1992 SO2
nonattainment status modestly reduced SO2 concentrations in the post-period.
7. Interpretation
The goal of the CAAAs is to bring all counties into compliance with the NAAQS. The EPA’sprimary tool to achieve this goal is its ability to designate counties nonattainment and strictlyregulate emitters in these areas. This paper has documented that by the late 1970s practically theentire US had achieved the SO2 standard. Its most surprising finding is that a substantial numberof counties with SO2 concentrations below the federal standard are designated SO2 nonattain-ment.Through extensive conversations with EPA staff, I received a number of explanations for this
seeming anomaly but none of them are completely satisfying.22 The most plausible explanationhas two parts. First, it is the states’ responsibility to petition to get a county redesignated fromnonattainment to attainment and this redesignation process is more difficult than might beexpected. For example, this process frequently requires states to develop a local air pollutionmodel that demonstrates that all areas within a county (not just the areas covered by themonitors) have SO2 concentrations below the federal threshold. These models are often necessaryeven when all the local pollution monitors record concentrations that are below the federalstandard. Importantly, these models are expensive to develop and states do not always have thenecessary resources.Second, the federal EPA staff suggested that they are receptive to requests to lessen the intensity
of regulatory oversight in nonattainment counties that do not exceed the NAAQS. It may beeasier to lessen the regulatory burden with these requests than by following all of the stepsinvolved in obtaining a formal redesignation to attainment status.Consequently, the stringency of regulations aimed at reducing a county’s SO2 pollution may
depend on a county’s nonattainment designation and whether its ambient SO2 concentrationexceeds the federal standards. If this is the case, this paper’s mixed results on the effectiveness ofthe SO2 nonattainment designation at reducing SO2 concentrations may not be surprising.However, it must be emphasized that this is a qualitative explanation of the results and cannotreadily be subject to a rigorous test.
ARTICLE IN PRESS
22This section draws on conversations with Matthew Witosky of the EPA Region 6 office and a number of people at
the national EPA office.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611608
8. Conclusion
This study has used regulation categories specified by the CAAAs to empirically test whetherthis legislation contributed to the dramatic 80% decline in sulfur dioxide air pollution thatoccurred over the last 30 years. Under this legislation, the EPA annually assigns every county anonattainment or attainment designation. Stricter regulations apply in the nonattainmentcounties. Two primary findings are that the statutory selection rule that should have determinednonattainment status is not followed and that there are important observable differences betweennonattainment and attainment counties. These findings make the evaluation problem especiallydifficult.Linear regression and propensity score matching are used to control for the likely confounding
caused by these differences. Notably, the matching appears to be especially useful when theobservable differences between the nonattainment and attainment counties are greatest. Theresults indicate that SO2 nonattainment status is associated with modest reductions in SO2 airpollution, but a null hypothesis of zero effect generally cannot be rejected.23 In summary, thepaper finds that the SO2 nonattainment designation played a minor role in the dramatic reductionof SO2 concentrations over the last three decades.The cause of the dramatic reduction in ambient SO2 remains a mystery. The most likely
candidates are other features of the CAAAs or state and local regulations. In support of the latterpossibility, there were large declines in SO2 concentrations that pre-dated the implementation ofthe 1970 CAAAs in a sample of 10 counties with monitors from 1969 to 1971. However, thisextremely small sample may not be representative of the rest of the country. Overall, there is littleevidence available to support or contradict alternative explanations.
Acknowledgments
The careful comments of Joseph Herriges and three anonymous reviewers substantiallyimproved this paper. Spencer Banzhaf, Gib Metcalf, and the participants at the NBER conference‘‘Empirical Advances in Environmental Economics’’ provided numerous insightful remarks. Ithank Katherine Ozment for many informative conversations. Matthew Witosky of the EPARegion 6 Office deserves special thanks for enlightening conversations about SO2 nonattainmentpolicies. Justin Gallagher, Paulette Kamenecka, and William Young provided outstandingresearch assistance.
Appendix
Table 6 reports the parameter estimates and standard errors form the control variables in thecolumn (4) specifications of panel (A) of Tables 3,4 and 5 when the dependent variables are the1980–1977, 1986–1983, and 1992–1989 SO2 concentrations, respectively.
ARTICLE IN PRESS
23This contrasts sharply with recent research that finds that the TSPs and ozone nonattainment designations are
associated with reductions in the ambient concentrations of these pollutants [6,7,16].
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 609
ARTICLE IN PRESS
Table 6
Estimated effects of determinants of changes in SO2 concentrations
Dependent variable 1980–1977 1986–1983 1992–1989
SO2 concentration SO2 concentration SO2 concentration
(1) (2) (3)
Mean SO2 t-2 �0.300 �0.780 0.174
(0.145) (0.538) (0.299)
Mean SO2 t-3 0.025 0.919 �0.818(0.174) (0.499) (0.302)
(Mean SO2 t-1)2 — �39.2 �76.6— (10.6) (24.7)
(Mean SO2 t-2)2 — 84.0 0.36
— (42.9) (19.13)
(Mean SO2 t-3)2 — �72.9 97.2
— (37.9) (24.5)
(Mean SO2 t-1)3 — 1078.4 2534.3
— (358.8) (921.5)
(Mean SO2 t-2)3 — �2532.2 �752.6— (1011.8) (514.6)
(Mean SO2 t-3)3 — 1939.9 �1935.2— (827.7) (454.5)
Maximum 1h t-1 �0.004 �0.0020 0.0168
(0.007) (0.0059) (0.0055)
Maximum 3h t-1 0.004 0.0071 �0.0279(0.010) (0.0099) (0.0078)
Maximum 24 h t-1 �0.021 0.0008 0.0535
(0.005) (0.0238) (0.0176)
(Maximum 1h t-1)2 — 0.0053 �0.0186— (0.0054) (0.0072)
(Maximum 3h t-1)2 — �0.0192 0.0328
— (0.0138) (0.0123)
(Maximum 24 h t-1)2 — 0.1012 �0.1949— (0.1073) (0.0665)
CO nonattainment t-1 0.0005 �0.0001 0.0005
(0.0011) (0.0005) (0.0003)
TSPs nonattainment t-1 �0.0001 �0.0002 0.00001
(0.0012) (0.0005) (0.00035)
Ozone nonattainment t-1 �0.0045 0.0006 0.0000041
(0.0020) (0.0004) (0.0003164)
Per capita income 0.00000078 0.000000052 0.000000018
(0.00000053) (0.000000077) (0.000000052)
Mean hourly wage 0.0000000086 �0.00000013 �0.00000010(0.0000003900) (0.00000010) (0.00000006)
Total employment 0.0000057 0.00000080 0.00000205
(0.0000099) (0.00000076) (0.00000083)
Total population �0.0000000032 �0.00000000051 �0.00000000122(0.0000000050) (0.00000000039) (0.00000000045)
Notes: The entries are the parameter estimates and heteroskedastic-consistent standard errors for the covariates from
the column (4) specifications in Tables 3A, 4A and 5A, respectively. See the notes to those tables.
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611610
References
[1] J.D. Angrist, Using social security data on military applicants to estimate the effect of military service earnings,
Econometrica 66 (2) (1998) 249–288.
[2] J.D. Angrist, V. Lavy, Does teacher training affect pupil learning? Evidence from matched comparisons in
Jerusalem public schools, NBER Working Paper 6781, 1998.
[3] R. Becker, V. Henderson, Effects of air quality regulations on polluting industries, J. Polit. Econ. 108 (2000)
379–421.
[4] D. Card, D. Sullivan, Estimating the effect of subsidized training on movements in and out of employment,
Econometrica 56 (3) (1988) 497–530.
[5] K. Chay, C. Dobkin, M. Greenstone, Air quality, adult mortality, and the clean air act of 1970, J. Risk
Uncertainty 27 (3) (2003) 279–300.
[6] K. Chay, M. Greenstone, Air quality, infant mortality, and the clean air act of 1970, NBERWorking Paper 10053,
2003.
[7] K. Chay, M. Greenstone, Does air quality matter? Evidence from the housing market, MIT Mimeograph, 2004.
[8] K. Chay, M. Greenstone, The impact of air pollution on infant mortality: evidence from geographic variation in
pollution shocks induced by a recession, Quart. J. Econ. 118 (3) (2003) 1121–1167.
[9] R.H. Dehejia, S. Wahba, Propensity score-matching methods for nonexperimental causal studies, Rev. Econ.
Statist. 84 (1) (2002) 151–161.
[10] D.W. Dockery, et al., An association between air pollution and mortality in six US cities, New England J. Med.
329 (24) (1993) 1753–1759.
[11] I.M. Goklany, Clearing the Air: The Real Story of the War on Air Pollution, The Cato Institute, Washington, DC,
1999.
[12] M. Greenstone, The impacts of environmental regulations on industrial activity: evidence from the 1970 and 1977
clean air act amendments and the census of manufacturers, J. Polit. Economy 110 (6) (2002) 1175–1219.
[13] J.J. Heckman, H. Ichimura, P. Todd, Matching as an econometric evaluation estimator, Rev. Econ. Stud. 65 (2)
(1998) 261–294.
[14] J.J. Heckman, Micro data, heterogeneity, and the evaluation of public policy: nobel lecture, J. Polit. Economy 109
(4) (2001) 673–748.
[15] J.J. Heckman, R. Robb, Alternative methods for evaluating the impact of interventions, in: J.J. Heckman, B.
Singer (Eds.), Longitudinal Analysis of Labor Market Data, Econometric Society Monograph, Vol. 10,
Cambridge University Press, Cambridge, UK, 1985.
[16] J.V. Henderson, Effects of air quality regulation, Amer. Econ. Rev. 86 (1996) 789–813.
[17] L.B. Lave, G.S. Omenn, Clearing the Air: Reforming the Clean Air Act, The Brookings Institution, Washington,
DC, 1981.
[18] R.A. Liroff, Reforming Air Pollution Regulations: The Toil and Trouble of EPA’S Bubble, The Conservation
Foundation, Washington, DC, 1986.
[19] M.R. Ransom, C.A. Pope III, External health costs of a steel mill, Contemporary Econ. Pol. 8 (1995) 86–97.
[20] P. Rosenbaum, D. Rubin, The central role of the propensity score in observational studies for causal effects,
Biometrika 70 (1983) 41–55.
[21] P. Rosenbaum, D. Rubin, Reducing bias in observational studies using subclassification on the propensity score,
J. Amer. Statist. Assoc. 79 (1984) 516–524.
[22] D. Rubin, Assignment to a treatment group on the basis of a covariate, J. Educ. Statist. 2 (1) (1977) 1–26.
[23] US Environmental Protection Agency, Office of Air Quality Planning and Standards, National Air Pollutant
Emission Trends, 1900–1994, US Environmental Protection Agency, Research, Triangle Park, NC, 1995.
[24] US Environmental Protection Agency, The benefits and costs of the clean air act, 1970–1990, Report to Congress
Mimeograph, 1997.
ARTICLE IN PRESS
M. Greenstone / Journal of Environmental Economics and Management 47 (2004) 585–611 611
MIT CENTER FOR ENERGY AND ENVIRONMENTAL POLICY RESEARCH REPRINT SERIES
CEEPR Reprints are available free of charge (limited quantities.) Order from MIT Center for Energy and Environmental Policy Research, E40-428, Cambridge, MA 02139-4307.
185. Do Markets Reduce Costs? Assessing the Impact of Regulatory Restructur-ing on US Electric Generation Effi ciency, Kira R. Fabrizio, Nancy L. Rose and Catherine D. Wolfram, AER, Vol. 97, No. 4, pp. 1250-1277, (2007)
186. Retail Electricity Competition, Paul Joskow and Jean Tirole, RAND Journal of Economics, Vol. 37, No. 4, pp. 799 - 815, (2006)
187. Reliability and Competitive Electricity Markets, Paul Joskow and Jean Tirole, RAND Journal of Economics, Vol. 38, No. 1, pp. 60-84, (2007)
188. The Eff ect of Variability in Industrial Emissions on Ozone Formation in Houston, Texas, Mort Webster, Junsang Nam, Yosuke Kimura, Harvey Jeff ries, William Vizuete and David T. Allen, Atmospheric Environment, Vol 41, No. 40, pp. 9580-9593, (2007)
189. A Simple Auction Mechanism for the Optimal Allocation of the Commons, Juan-Pablo Montero, AER, Vol. 98, No. 1, pp. 496-518, (2008)
190. Capacity Payments in Imperfect Electricity Markets: Need and Design, Paul L. Joskow, Utilities Policy, Vol. 16, No. 3, pp. 159-170, (2008)
191. The Effi ciency and Robustness of Allowance Banking in the U.S. Acid Rain Program, A. Denny Ellerman andJuan-Pablo Montero, The Energy Journal, Vol. 28, No. 4, pp. 67-92, (2007)
192. Reductions in Ozone Concentrations due to Controls on Variability in Indus-trial Flare Emissions in Houston, Texas, Junsang Nam, Mort Webster, Yosuke Kimura, Harvey Jeff ries, William Vizuete and David T. Allen, Atmospheric Environment, Vol. 42, pp. 4198-4211, (2008)
193. Competitive Electricity Markets and Investment in New Generating Capacity, Paul Joskow, in The New Energy Paradigm, D. Helm (Editor), Oxford University Press, pp. 76-122, (2007)
194. International Market Integration for Natural Gas? A Cointegration Analysis of Prices in Europe, North America and Japan, Boriss Siliverstovs, Guillaume L'Hégaret, Anne Neumann and Christian von Hirschhausen, Energy Economics, Vol. 27, No. 4, pp. 603-615, (2005)
195. Managing a Portfolio of Real Options: Sequential Exploration of Dependent Prospects, James L. Smith and Rex Thompson, Energy Journal, Vol. 29, Special Issue, (2008)
196. Did the Clean Air Act Cause the Remarkable Decline in Sulfur Dioxide Concentrations?, Michael Greenstone, Journal of Environmental Economics and Management, Vol. 47, No. 3, pp. 585-611, (2004)
Massachusetts Institute of Technology
Center for Energy and Environm
ental Policy ResearchO
ne Am
herst Street (E40-428)C
amb
ridge, Massachusetts 02139-4307